Multi-view ear recognition by patrial least square discrimination

In this work, partial least square discrimination (PLSD) based multi-view ear recognition is first time well investigated. In order to study the actual classification performance of partial least square representation, instead of the traditional recognition style - using partial least square to do feature extraction and then taking diverse classifiers for classification, we do directly take partial least square regression for test samples classification. In addition, one modern feature extraction technique - random projection is discussed its effect on the performance for multi-view ear recognition under different ear dataset. The experimental results and the comparisons show that, PLSD can get rather good with stable recognition performance even under different multi-view ear dataset. This indicates us for multi-view ear recognition scenario, PLSD can be regarded as a benchmark for different recognition methods under different multi-view dataset.